Provide bn_mul_comba6().

This allows for fast multiplication of two 6 word arrays.

ok tb@

2 files changed, 63 insertions, 2 deletions

diff --git a/src/lib/libcrypto/bn/bn_local.h b/src/lib/libcrypto/bn/bn_local.h
index af280ebfce..39e996bca7 100644
--- a/src/lib/libcrypto/bn/bn_local.h
+++ b/src/lib/libcrypto/bn/bn_local.h
@@ -1,4 +1,4 @@
-/* $OpenBSD: bn_local.h,v 1.52 2025/08/05 15:01:13 jsing Exp $ */
+/* $OpenBSD: bn_local.h,v 1.53 2025/08/05 15:06:13 jsing Exp $ */
 /* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com)
  * All rights reserved.
  *
@@ -241,6 +241,7 @@ BN_ULONG bn_sub(BN_ULONG *r, int r_len, const BN_ULONG *a, int a_len,
 
 void bn_mul_normal(BN_ULONG *r, BN_ULONG *a, int na, BN_ULONG *b, int nb);
 void bn_mul_comba4(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b);
+void bn_mul_comba6(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b);
 void bn_mul_comba8(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b);
 
 void bn_sqr_comba4(BN_ULONG *r, const BN_ULONG *a);
diff --git a/src/lib/libcrypto/bn/bn_mul.c b/src/lib/libcrypto/bn/bn_mul.c
index ebf34bb413..70f6534b8f 100644
--- a/src/lib/libcrypto/bn/bn_mul.c
+++ b/src/lib/libcrypto/bn/bn_mul.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: bn_mul.c,v 1.41 2025/08/05 15:01:13 jsing Exp $ */
+/* $OpenBSD: bn_mul.c,v 1.42 2025/08/05 15:06:13 jsing Exp $ */
 /* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com)
  * All rights reserved.
  *
@@ -104,6 +104,66 @@ bn_mul_comba4(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b)
 #endif
 
 /*
+ * bn_mul_comba6() computes r[] = a[] * b[] using Comba multiplication
+ * (https://everything2.com/title/Comba+multiplication), where a and b are both
+ * six word arrays, producing a 12 word array result.
+ */
+#ifndef HAVE_BN_MUL_COMBA6
+void
+bn_mul_comba6(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *b)
+{
+        BN_ULONG c0, c1, c2;
+
+        bn_mulw_addtw(a[0], b[0],  0,  0,  0, &c2, &c1, &r[0]);
+
+        bn_mulw_addtw(a[0], b[1],  0, c2, c1, &c2, &c1, &c0);
+        bn_mulw_addtw(a[1], b[0], c2, c1, c0, &c2, &c1, &r[1]);
+
+        bn_mulw_addtw(a[2], b[0],  0, c2, c1, &c2, &c1, &c0);
+        bn_mulw_addtw(a[1], b[1], c2, c1, c0, &c2, &c1, &c0);
+        bn_mulw_addtw(a[0], b[2], c2, c1, c0, &c2, &c1, &r[2]);
+
+        bn_mulw_addtw(a[0], b[3],  0, c2, c1, &c2, &c1, &c0);
+        bn_mulw_addtw(a[1], b[2], c2, c1, c0, &c2, &c1, &c0);
+        bn_mulw_addtw(a[2], b[1], c2, c1, c0, &c2, &c1, &c0);
+        bn_mulw_addtw(a[3], b[0], c2, c1, c0, &c2, &c1, &r[3]);
+
+        bn_mulw_addtw(a[4], b[0],  0, c2, c1, &c2, &c1, &c0);
+        bn_mulw_addtw(a[3], b[1], c2, c1, c0, &c2, &c1, &c0);
+        bn_mulw_addtw(a[2], b[2], c2, c1, c0, &c2, &c1, &c0);
+        bn_mulw_addtw(a[1], b[3], c2, c1, c0, &c2, &c1, &c0);
+        bn_mulw_addtw(a[0], b[4], c2, c1, c0, &c2, &c1, &r[4]);
+
+        bn_mulw_addtw(a[0], b[5],  0, c2, c1, &c2, &c1, &c0);
+        bn_mulw_addtw(a[1], b[4], c2, c1, c0, &c2, &c1, &c0);
+        bn_mulw_addtw(a[2], b[3], c2, c1, c0, &c2, &c1, &c0);
+        bn_mulw_addtw(a[3], b[2], c2, c1, c0, &c2, &c1, &c0);
+        bn_mulw_addtw(a[4], b[1], c2, c1, c0, &c2, &c1, &c0);
+        bn_mulw_addtw(a[5], b[0], c2, c1, c0, &c2, &c1, &r[5]);
+
+        bn_mulw_addtw(a[5], b[1],  0, c2, c1, &c2, &c1, &c0);
+        bn_mulw_addtw(a[4], b[2], c2, c1, c0, &c2, &c1, &c0);
+        bn_mulw_addtw(a[3], b[3], c2, c1, c0, &c2, &c1, &c0);
+        bn_mulw_addtw(a[2], b[4], c2, c1, c0, &c2, &c1, &c0);
+        bn_mulw_addtw(a[1], b[5], c2, c1, c0, &c2, &c1, &r[6]);
+
+        bn_mulw_addtw(a[2], b[5],  0, c2, c1, &c2, &c1, &c0);
+        bn_mulw_addtw(a[3], b[4], c2, c1, c0, &c2, &c1, &c0);
+        bn_mulw_addtw(a[4], b[3], c2, c1, c0, &c2, &c1, &c0);
+        bn_mulw_addtw(a[5], b[2], c2, c1, c0, &c2, &c1, &r[7]);
+
+        bn_mulw_addtw(a[5], b[3],  0, c2, c1, &c2, &c1, &c0);
+        bn_mulw_addtw(a[4], b[4], c2, c1, c0, &c2, &c1, &c0);
+        bn_mulw_addtw(a[3], b[5], c2, c1, c0, &c2, &c1, &r[8]);
+
+        bn_mulw_addtw(a[4], b[5],  0, c2, c1, &c2, &c1, &c0);
+        bn_mulw_addtw(a[5], b[4], c2, c1, c0, &c2, &c1, &r[9]);
+
+        bn_mulw_addtw(a[5], b[5],  0, c2, c1, &c2, &r[11], &r[10]);
+}
+#endif
+
+/*
  * bn_mul_comba8() computes r[] = a[] * b[] using Comba multiplication
  * (https://everything2.com/title/Comba+multiplication), where a and b are both
  * eight word arrays, producing a 16 word array result.